Swarm and Evolutionary Computation 44 (2019) 113–129 Contents lists available at ScienceDirect Swarm and Evolutionary Computation journal homepage: www.elsevier.com/locate/swevo Make robots be bats: specializing robotic swarms to the Bat algorithm Patricia Suárez a,AndrésIglesiasa,b,*,AkemiGálveza,b a Department of Applied Mathematics and Computational Sciences, E.T.S.I. Caminos, Canales y Puertos, University of Cantabria, Avda. de los Castros, s/n, 39005, Santander, Spain b Department of Information Science, Faculty of Sciences, Toho University, 2-2-1 Miyama 274-8510, Funabashi, Japan ARTICLE INFO ABSTRACT Keywords: Bat algorithm is a powerful nature-inspired swarm intelligence method proposed by Prof. Xin-She Yang in 2010, Swarm computation with remarkable applications in industrial and scientific domains. However, to the best of authors’ knowledge, Swarm robotics this algorithm has never been applied so far in the context of swarm robotics. With the aim to fill this gap, this Bat algorithm paper introduces the first practical implementation of the bat algorithm in swarm robotics. Our implementation Unknown target location Behavioral patterns is performed at two levels: a physical level, where we design and build a real robotic prototype; and a computa- tional level, where we develop a robotic simulation framework. A very important feature of our implementation is its high specialization: all (physical and logical) components are fully optimized to replicate the most relevant features of the real microbats and the bat algorithm as faithfully as possible. Our implementation has been tested by its application to the problem of finding a target location within unknown static indoor 3D environments. Our experimental results show that the behavioral patterns observed in the real and the simulated robotic swarms are very similar. This makes our robotic swarm implementation an ideal tool to explore the potential and limitations of the bat algorithm for real-world practical applications and their computer simulations. 1. Introduction In this context, one of the most exciting breakthroughs in artifi- cial intelligence during the last decades is the adoption and subse- 1.1. Swarm intelligence quent popularization of this collective intelligence arising from a col- lection of simple, unsophisticated, and generally homogeneous agents For many years, the concept of intelligence was generally perceived collaborating together to solve a complex problem. This field, globally to be an exclusive attribute of highly sophisticated individuals. Not sur- known as swarm intelligence [5,6], is overcoming the traditional math- prisingly, this was also the approach taken in the early days of the ematical approaches for solving optimization problems and laying the artificial intelligence field. However, nature provides several examples foundations for a new computational paradigm: the swarm computation of groups of animals that exhibit a more sophisticated social behavior [13,27]. Under this new paradigm, there is no centralized intelligence than what would be possible through the simple aggregation of individ- controlling the swarm, taking decisions, and dictating how the swarm ual behavioral patterns. As a result, the social group is able to carry out units should behave. Instead, local and (at some extent) random inter- complex tasks that their individual members cannot intend for. Take, actions between simple agents lead to the emergence of global sophisti- for instance, the colonies of social insects (ants, termites, bees, fire- cated “intelligent” behaviors, unknown to the individual agents. flies), and other animal formations (fish schooling, bird flocking, animal Nowadays, swarm intelligence is attracting increasing attention herding, and so on). They come up with astonishingly complex social from researchers and practitioners owing to its potential applications behaviors from the combined efforts of individuals with extremely lim- to many problems. For instance, military and civil applications related ited intelligence. Amazingly, this complex collective behavior emerges to the control of unmanned vehicles have been described in Refs. from a small set of simple behavioral rules exploiting only low-level [45,46,54]. Many other applications can also been found in the sci- interactions between individuals and with the environment (stigmergy) using decentralized control and self-organization. * Corresponding author. Department of Applied Mathematics and Computational Sciences, E.T.S.I. Caminos, Canales y Puertos, University of Cantabria, Avda. de los Castros, s/n, 39005, Santander, Spain. E-mail address: [email protected] (A. Iglesias). URL: http://personales.unican.es/iglesias (A. Iglesias). https://doi.org/10.1016/j.swevo.2018.01.005 Received 24 July 2017; Received in revised form 20 November 2017; Accepted 9 January 2018 Available online 14 February 2018 2210-6502/© 2018 Elsevier B.V. All rights reserved. P. Suárez et al. Swarm and Evolutionary Computation 44 (2019) 113–129 entific literature; see Refs. [5,19,63] for several illustrative examples in microbats (see Section 3 for details). In our experience, the bat algo- different fields. The interested reader is also referred to [13,27]fora rithm has shown to be a very effective method to solve complex mul- comprehensive overview about the field of swarm intelligence, its his- timodal nonlinear continuous optimization problems involving a large tory, main techniques, and applications. number of variables, such as data fitting through free-form paramet- ric curves [20,21,23]andsurfaces[22], and multi-objective problems 1.2. Swarm robotics [59]. On the other hand, the fundamental principle of the bat algorithm (namely, the echolocation through ultrasounds) along with some of its One of most relevant applications of swarm intelligence is robotics. most important features and parameters (operating frequency, operat- Several scientific papers and research projects have shown that self- ing time, number of cycles per bust, accuracy range, traveling range of organizing swarm robots can potentially accomplish complex tasks and pulses, and so on) can readily be reproduced with current hardware (see thus replace sophisticated and expensive robots by simple inexpen- Section 4 for details). This makes the bat algorithm an excellent method sive drones, a research subfield usually referred to as swarm robotics for potential applications in swarm robotics. Furthermore, some recent [3,14,64]. The reader is kindly referred to [56] for some illustrative papers have reported successful applications of the bat algorithm to early examples and applications of swarm robotics; see also [50]fora some problems in robotics [15,16]. However, these works are mostly more updated survey on recent advances in the field. As remarked by focused on the problem of parameter tuning of PI-controllers for opti- several authors [1,6], swarm robotic systems offer several interesting mal positioning of robotic arms for safety and functionality, and have advantages, such as: no direct relationship with the area of swarm robotics. Clearly, the application of bat algorithm to swarm robotics is a very promising yet • Improved performance by parallelization: swarm intelligence systems unexplored field. In this paper we are aimed at filling this gap. Conse- are very well suited for parallelization, because the swarm mem- quently, this is the algorithm used in this paper. bers can perform different actions at different locations simultane- ously.Thisfeaturemakestheswarmmoreflexibleandefficientfor 1.4. Main contributions and structure of the paper complex tasks, as individual robots (or groups of them) can solve different parts of a complex task independently. Despite of its remarkable features, to the best of our knowledge, the • Task enablement: groups of robots can do certain tasks that are bat algorithm has never been applied so far in the context of swarm impossible or very difficult for a single robot (e.g., collective trans- robotics. Aimed at filling this gap, a previous paper in Ref. [49]pre- port of too heavy items, dynamic target tracking, cooperative envi- sented some initial ideas about the application of the bat algorithm to ronment monitoring, autonomous surveillance of large areas). swarm robotics. However, that paper was based on very preliminary • Scalability: inclusion of new robots into a swarm does not require work and strongly affected by limitations of space. This paper is a sub- reprogramming the whole swarm. Furthermore, because interac- stantial extension in several ways. The main contributions of this paper tions between robots involve only neighboring individuals, the total are: number of interactions within the system does not increase dramat- ically by adding new units. 1. First and utmost contribution of this work is the first practical imple- • Distributed sensing and action: a swarm of simple interconnected mentation of the bat algorithm for swarm robotics. To the best of our mobile robots deployed throughout a large search space possesses knowledge, no previous application of the bat algorithm to swarm greater exploratory capacity and a wider range of sensing than a robotics has been reported so far in the literature. sophisticated robot. This makes the swarm much more effective 2. Our implementation is performed at two levels: we introduce both in several tasks: exploration and navigation (e.g., in disaster res- a physical robotic prototype (described in Section 4.1)andacomputa- cue missions), nanorobotics-based manufacturing, microbotics